Part 2: Workflow map optimization by using multiobjective algorithms Determine the optimum process paths for the selected criteria by application of multiobjective optimization algorithms 4/12/2012 IBM Lukasz Osuszek About the author: Abstract: This paper covers extending Process Definition Language (XPDL), Workflow map and application of multi-objective optimization algorithms to enable fully automated optimization of the Workflow map. The mathematical model of the business process may be subject to specific multi-criteria optimization algorithms Lukasz Osuszek is with IBM ECM technology pre-sales, architect and technical support for IBM FileNet P8. He has eight years experience in ECM area, and four years experience in IBM Polish Software Group. Reach out to him at lukasz.osuszek@pl.ibm.com
Part 1: Introduction to BPM code optimization Introduction In the previous article we introduced BPM workflow improvement by adopting code optimization techniques. In this article we look at multi-objective workflow optimization and tangible values it might brought to Business Process Management. Today, there are several techniques for modeling business processes which enable mapping existing processes of the organization, and allow for creating new ones in order to meet growing market demand. One of the process maps - created in IBM Case Manager (ICM) - is rendered below in figure 1. For each company and organization, business processes and the related decisions are the key element which provides the momentum for their operations and determine their competitiveness. The management of workflow and information within process paths has a major impact on the speed, flexibility and quality of decision-making processes. This is why the acceleration and optimization of processes is decisive for the success of any organization. Figure 1. Business process map in IBM Case Manager Processes involve people, systems and information. The maximum efficiency is possible only if all of these elements interoperate in an automated environment. Note also that optimized processes enable a faster response to the changing market situation and to new customers demands while guaranteeing compliance with applicable regulations. In short, 2
better processes contribute towards continuous improvement of the efficiency of company s operations, and therefore, allow gaining competitive advantage in the industry. One of the factors which make the Business Process Management systems increasingly popular is the tracking, analysis and simulation of processes. With the monitoring of work progress, with in-depth analysis of current and historical processes, and with the verification of changes to processes prior to their implementation in a production system, these tools guarantee more accurate business decisions. Additionally, they enable fast implementation of best business practices, and reduce the total cost of system ownership with reusable process definitions. The aforementioned advantages of BPM are quite widely spoken of, but not enough attention is paid to the optimization problem. Today s tools offer the functionality of business process optimization. However, expert knowledge is required to use them efficiently. With their experience backed up by software-based simulations, the consultants who operate such tools are able to specify the way of optimization of the process concerned. The weakness of the BPM description language is the inadequate description model. Only few description models of business processes enable the use of process map optimization methods and analyses to improve the timeliness (accelerate), to ensure the optimum use of resources, or to save money. The objective of this paper is to show the ability to extend the business process description model in order to enable a more sophisticated, multi-objective optimization. Idea of multi-objective optimization could be easily adapted to real use cases by combination of BVA (Business Value Assessment) and mathematical models. BVA may be used to perform an in-depth cost analysis of each business process component. Such a descriptive model provides additional analysis and optimization possibilities. This data could be easily used to mathematical optimization algorithms. It is a powerful tool that can suggest how to restructure the company or customize the existing business process so as to make it more optimal. 3
Current fields of studies Several models have been developed to enable description of business processes using mathematical models. The most popular include Integration Definition (IDEF), Computer Integrated Manufacturing Open System Architecture (CIM-OSA), Object-Oriented Modeling, and the highly popular Petri Nets. These standards were used to develop many tools for business process modeling (ARIS, FirstStep, PrimeObject, etc.). Zakarian [1] integrated the Fuzzy-rule-based Reasoning Approach with IDEF in order to extend the quantitative analysis of the process model. Grigori [2] proposed the Business Process Intelligence, a tool which uses data mining methods for the analysis of business processes. Multiple algorithms were developed to enable optimization of business problems in the area of logistics Yu and Li [3], as most of business models (including Business Process Modeling Notation) are insufficient from the point of view of the multi-criteria analysis. McKay and Radnor [4] presented a model for the description of business processes which, however, did not include any formal optimization methods. Most scientific studies on business process optimization focus on selecting the appropriate process model, or on one-dimensional optimization (Hofacker and Vetschera [5]) which is unsatisfactory. Multi-objective (multi-criteria) optimization Generally, there are the following optimization types: Single-criterion optimization: if the ideal state is required to be reached for a single evaluation criterion; Multi-criteria optimization (vector optimization, poly-optimization): if reaching the ideal state depends on multiple evaluation criteria. A large number of criteria for the evaluation of the ideal state often results in contradictions between them. This means that the solution looked for does not reach the extreme values of all criteria considered separately. Instead, it provides some kind of compromise between them. Therefore, the poly-optimization problem consists primarily in defining that compromise. In many cases, the heuristic knowledge about the optimized process allows for specifying another, substitute criterion for searching the compromise solution. In formal terms, poly-optimization may be specified as follows: Let X = {xl}, l = 1, 2,..., N be a vector of decision variables considered as independent. Let F = {fi}, i = 1, 2,..., M be a set of criteria (functions) for evaluating solutions when looking for the compromise. Let the following restrictions be imposed on the values of the solutions: inequality restrictions: G = {gk}, k = 1, 2,..., K, with: gk (X) 0; equality restrictions: H = {hj}, j = 1, 2,..., J, with: hj (X) = 0; The objective of poly-optimization is to reach a solution which meets the following condition: 4
min F(X) ={ f1(x), f2 (X),..., fl (X)} If maximization of an fl * function is required, an auxiliary criterion may be introduced in accordance with the following formula: min fl (X) = -max fl* (-X) Figure 2. Business process design with activities and resources In most cases, the business process description model includes activities and resources (figure 2). The activities are supposed to enable meeting the objective of the business process. The two sets of resources showed on the figure (Iglob and Oglob) are, respectively, the initiating resources available at the beginning of the process, and the output resources resulting from the performance of the process. There are two categories of resources flowing through the entire map of the business process: Physical resources (e.g., process participants) and Information resources. Business process optimization involves specification of the criteria to be optimized. Usually, these will be costs and process duration. 5
The literature provides numerous examples of ready-to-use algorithms for multi-criteria optimization of various types of problems. These algorithms may use any criteria through the application of the corresponding mathematical model. The unambiguous conclusion from the analysis of existing multi-criteria optimization algorithms is that the model for the XPDL description of business processes is simple and may be extended with additional information enabling the construction of a more accurate mathematical model. This conclusion is applicable to the XPDL supported by P8 BPM, because as most BPM models, the XPDL model can be represented in BPMN. Let us consider the following example of a business process: Figure 3. Travel Agency process of holidays offering Figure 3 renders generic model of the operation of a virtual travel agency. The input (initiating) data is the customer s guidelines as to the details of a trip, and the maximum price the customer may pay. Then, the travel agent performs a series of interrelated actions (activities) which compose a business process aimed at the delivery of a proposal compliant with the expectations. Let us assume extension of the standard BPM model describing the business process with additional information on the costs and duration of each action (business step). Additionally, detailed possible actions are specified for each activity. A similar approach was included in IBM Case Manager, the latest tool for describing dynamic business processes. Table 1 describes the business process of finding the appropriate trip proposal. 6
Table 1. Process map as a set of process elements, cost and duration Object name Process element Alternatives Cost Duration Travel details Input resource - - - Price limit Input resource - - - Browse 1. Search from brochures 2 9 Activity pre-booked packages 2. Search company intranet 7 5 Explore travel options 1. Browse past cases 4 8 Activity 2. Explore new options 6 6 Check availability Activity 1. Via intranet/e-mail 10 1 7 2. Via phone/post 5 7 1. Use specific software 11 2 2. Combine options manually 5 6 Create tailored package Activity Holiday proposals Output resource - - - Payment details Output resource - - - The customer who enters the travel agency reports his/her request to prepare a proposal of holiday in a specific localization. He/she also specifies the price limit to be complied with. To reply, the travel agent may browse through the previously prepared trip packages, or search through the entire proposal database in order to prepare a customized offer. There are two ways (activities) to browse through previously prepared packages: checking the information brochures or searching through the company s intranet database. If the agent decides to perform a more in-depth exploration of the proposal database, he/she may check the past cases or verify new options. Once the details of the choice of the trip are determined, the agent checks the availability of the proposal (through intranet/e-mail or phone/mail). The last step of the proposal construction process is the presentation of a customized trip package to the customer. To do so, the agent may use dedicated tools, or he may create the proposal manually. Each of the selected actions involves the corresponding cost and duration of the activity. Description of the business process extended in that manner enables creation of a mathematical model which may be optimized with known multi-criteria optimization algorithms. The use of Non-Dominated Sorting Genetic Algorithm II (NSGA2), Strength Pareto Evolutionary Algorithm II (SPEA2) or Multi-Objective Particle Swarm Optimization (MOPSO) allows looking for process map variants optimized for the selected criteria. The process diagrams below present the paths optimized for the duration or for the costs generated by the process. In this experiment SPEA2 algorithm was choose as a core of multi-objective optimization. The Strength Pareto Evolutionary Algorithm (SPEA) [6] is a relatively recent technique for finding or approximating the Pareto-optimal set for multi-objective optimization problems. SPEA has shown very good performance in comparison to other multi-objective evolutionary algorithms [7], and therefore it has been a point of reference in various recent investigations. In this experiment, an improved version, namely SPEA2, is applied, which incorporates in contrast to its predecessor a fine-grained fitness assignment strategy, a density estimation technique, and an enhanced archive truncation method. For the purposes of this experiment, a dedicated application was developed which optimizes the process map in function of the selected criterion by using SPEA2 algorithm.
If the process performance time is specified as the optimization criterion (Figure 5), the application analyzes possible combinations of activities in order to determine the shortest delivery path for the entire process. If the main criterion is cost (Figure 4) algorithm build the cheapest process map. Figure 4. Workflow path optimized for cost criterion. Figure 5. Workflow path optimized for time criterion. 8
For that purpose, we create an appropriate text file which includes the process map description: Figure 6. Description of process map With a proper structure which reflects business process: zero - a 8 8 ; b 7 7 one - c 1 2 ; d 3 3 two - e 5 5 ; f 9 9 Connections 0 1 ; 1 2 The name of the object, and the possible alternatives of activities together with the information on the costs and duration in the process. The tool provides an optimized process path for the selected criterion (or a conjunction of criteria). The SPEA2 Algorithm SPEA2 was designed to overcome the aforementioned problems. The overall algorithm is as follows: Algorithm 1 (SPEA2 Main Loop) Input: N (population size) N (archive size) T (maximum number of generations) Output: A (non-dominated set) Step 1: Initialization: Generate an initial population P 0 and create the empty archive (external set) 9 P. Set t = 0. 0 Step 2: Fitness assignment: Calculate fitness values of individuals in P t and P t Step 3: Environmental selection: Copy all non-dominated individuals in P t and Pt to P t 1. If size of P t 1 exceeds N then reduce Pt 1by means of the truncation operator, otherwise if size of Pt 1is less than N then fill Pt 1with dominated individuals in P t and P t Step 4: Termination: If t T or another stopping criterion is satisfied then set A to the set of decision vectors represented by the non-dominated individuals in P t 1. Stop. Step 5: Mating selection: Perform binary tournament selection with replacement on
Pt 1 in order to fill the mating pool Step 6: Variation: Apply recombination and mutation operators to the mating pool and set P to the resulting population. Increment generation counter (t = t + 1) and go to Step 2. t 1 In contrast to SPEA, SPEA2 uses a fine-grained fitness assignment strategy which incorporates density information. Furthermore, the archive size is fixed, i.e., whenever the number of non-dominated individuals is less than the predefined archive size, the archive is filled up by dominated individuals; with SPEA, the archive size may vary over time. In addition, the clustering technique, which is invoked when the non-dominated front exceeds the archive limit, has been replaced by an alternative truncation method which has similar features but does not loose boundary points. Finally, another difference to SPEA is that only members of the archive participate in the mating selection process Conclusions The approach aimed at expanding the business process description model focuses on the mathematical analysis of the business. The traditional approach towards business process optimization, offered by various IBM tools (e.g., Process Analyzer), focuses more on the aspect of actors (participants) of the process. This allows the experts to quickly find the system steps that may be performed in parallel, or to avoid any dead ends or redundant iterations. However, the proper use of such types of tools requires expert knowledge and methodic experience in the development of business paths. The extension of the BPM for business process description, as proposed in this article, enables a fully automated optimization of the business process. The mathematical model of the business process may be subject to a specific multi-criteria analysis in order to determine the optimum process paths for the selected criterion. Real life uses cases could be optimized by an in-depth analysis of individual tasks (time consumption, costs) comprising business processes makes it possible to identify those areas which may generate savings upon modification. Modification could be prompted by adopting multi-objective algorithms like SPEA2. Reader could also be interesting in further article. Next part introduces conversion of XPDL workflows into Petri Network Modeling Notation for optimization in category of time consumption. 10
Bibliography [1] Zakarian A., 2001. Analysis of process models: A fuzzy logic approach. The International Journal of Advanced Manufacturing Technology 17, 444-452. [2] Grigori D., Casati F., Castellanos M., Dayal U., Sayal M. and Shan M.C., 2004. Business Process Intelligence. Computers in Industry 53, 321-343. [3] Yu C-S. and Li H-L., 2000. A robust optimization model for stochastic logistic problems. International Journal of Production Economics 64, 385-397. [4] McKay A. and Radnor Z., 1998. A characterization of a business process. The International Journal of Operations and Production Management18 (9/10), 924-936. [5] Hofacker I. and Vetschera R., 2001. Algorithmical approaches to business process design. Computers & Operations Research 28, 1253-1275. [6] Zitzler, E. (1999). Evolutionary Algorithms for Multiobjective Optimization: Methods and Applications. Ph. D. thesis, Swiss Federal Institute of Technology (ETH) Zurich, Switzerland. TIK-Schriftenreihe Nr. 30, Diss ETH No. 13398, Shaker Verlag, Aachen, Germany [7] Zitzler, E., K. Deb, and L. Thiele (1999, December). Comparison of multiobjective evolutionary algorithms: Empirical results (revised version). Technical Report 70, Computer Engineering and Networks Laboratory (TIK), Swiss Federal Institute of Technology (ETH) Zurich, Gloriastrasse 35, CH-8092 Zurich, Switzerland. [8] K. Vergidis, A. Tiwari, B. Majeed Optimisation of Business Process Designs: An algorithmic approach with multiple objectives. [9] J. M. Pinto and I. E. Grossmann, Assignment and sequencing models for the scheduling of process systems, Ann. Oper. Res., vol. 81, pp. 433 466,1998 [10] F. Soliman, Optimum level of process mapping and least cost business process re-engineering, Int. J. Oper. Prod. Manage., vol. 18, no. 9/10, pp. 810 816, 1998 [11] Tiwari, K. Vergidis, and B. Majeed, Evolutionary multi-objective optimisation of business processes, in Proc. IEEE Congr. Evol. Comput.,Jul. 2006, pp. 3091 3097. [12] W. M. P. van der Aalst, A. H. M. ter Hofstede, and M. Weske, Business process management: A survey, in Lecture Notes Computer Sciences, Springer-Verlag, 2003, vol. 2678, pp. 1 12. [13] WIL M. P. VAN DER AALST ET AL, Pattern-Based Analysis of BPML and WSCI, 2004 Sample application for workflow cost optimization C:\2\cost\ cost_optimization.rar 11